Loading graph…

Ensemble Learning

Definition

Ensemble learning combines multiple models to produce one stronger prediction.

Instead of relying on one model:

\[\hat f_1(x)\]

an ensemble combines many models:

\[\hat f_1(x), \hat f_2(x), \ldots, \hat f_B(x)\]

Main Idea

Different models make different errors.

If their errors are not perfectly identical, averaging them can reduce variance and improve prediction stability.

For regression, an ensemble often predicts:

\[\hat f(x) = \frac{1}{B}\sum_{b=1}^{B}\hat f_b(x)\]

Common Ensemble Methods

Why It Helps

A single tree can be unstable.

Averaging many trees usually produces a smoother and more reliable prediction.

This is the main idea behind random forests.

Retail Example

For basket-size prediction, one tree may overreact to unusual large orders.

An ensemble averages many trees, reducing the effect of any single unstable split.

Strengths

  • Often improves predictive accuracy.
  • Reduces variance.
  • Works well with nonlinear data.
  • Useful for complex datasets.

Weaknesses

  • Less interpretable than a single model.
  • More computationally expensive.
  • Can be harder to explain in a report.

Exercises

  1. Why can averaging models reduce variance?
  2. Why is a random forest usually more stable than one decision tree?
  3. Give one disadvantage of ensemble methods.

See

Decision Trees

Random Forests

Bagging

25

25
Ready to start
Ensemble Learning
Session: 1 | Break: Short
Today: 0 sessions
Total: 0 sessions